aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
authorMuli Ben-Yehuda <muli@il.ibm.com>2007-07-21 17:11:04 +0200
committerLinus Torvalds <torvalds@woody.linux-foundation.org>2007-07-21 18:37:11 -0700
commit820a149705c2c2a37989554a4f4a34e3d0b0df1f (patch)
treed6f13abb1b2640b598907c00d8eb4d2c17743b68 /arch
parent7354b07595b2e43b75fe353fcf18e73eb0427c9b (diff)
downloadkernel_samsung_tuna-820a149705c2c2a37989554a4f4a34e3d0b0df1f.zip
kernel_samsung_tuna-820a149705c2c2a37989554a4f4a34e3d0b0df1f.tar.gz
kernel_samsung_tuna-820a149705c2c2a37989554a4f4a34e3d0b0df1f.tar.bz2
x86_64: Calgary - tighten up the bitmap locking
Currently the IOMMU table's lock protects both the bitmap and access to the hardware's TCE table. Access to the TCE table is synchronized through the bitmap; therefore, only hold the lock while modifying the bitmap. This gives a yummy 10-15% reduction in CPU utilization for netperf on a large SMP machine. Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'arch')
-rw-r--r--arch/x86_64/kernel/pci-calgary.c40
1 files changed, 17 insertions, 23 deletions
diff --git a/arch/x86_64/kernel/pci-calgary.c b/arch/x86_64/kernel/pci-calgary.c
index a2f9c53..fa50d6a 100644
--- a/arch/x86_64/kernel/pci-calgary.c
+++ b/arch/x86_64/kernel/pci-calgary.c
@@ -233,6 +233,7 @@ static void iommu_range_reserve(struct iommu_table *tbl,
unsigned long index;
unsigned long end;
unsigned long badbit;
+ unsigned long flags;
index = start_addr >> PAGE_SHIFT;
@@ -244,6 +245,8 @@ static void iommu_range_reserve(struct iommu_table *tbl,
if (end > tbl->it_size) /* don't go off the table */
end = tbl->it_size;
+ spin_lock_irqsave(&tbl->it_lock, flags);
+
badbit = verify_bit_range(tbl->it_map, 0, index, end);
if (badbit != ~0UL) {
if (printk_ratelimit())
@@ -253,15 +256,20 @@ static void iommu_range_reserve(struct iommu_table *tbl,
}
set_bit_string(tbl->it_map, index, npages);
+
+ spin_unlock_irqrestore(&tbl->it_lock, flags);
}
static unsigned long iommu_range_alloc(struct iommu_table *tbl,
unsigned int npages)
{
+ unsigned long flags;
unsigned long offset;
BUG_ON(npages == 0);
+ spin_lock_irqsave(&tbl->it_lock, flags);
+
offset = find_next_zero_string(tbl->it_map, tbl->it_hint,
tbl->it_size, npages);
if (offset == ~0UL) {
@@ -270,6 +278,7 @@ static unsigned long iommu_range_alloc(struct iommu_table *tbl,
tbl->it_size, npages);
if (offset == ~0UL) {
printk(KERN_WARNING "Calgary: IOMMU full.\n");
+ spin_unlock_irqrestore(&tbl->it_lock, flags);
if (panic_on_overflow)
panic("Calgary: fix the allocator.\n");
else
@@ -281,17 +290,17 @@ static unsigned long iommu_range_alloc(struct iommu_table *tbl,
tbl->it_hint = offset + npages;
BUG_ON(tbl->it_hint > tbl->it_size);
+ spin_unlock_irqrestore(&tbl->it_lock, flags);
+
return offset;
}
static dma_addr_t iommu_alloc(struct iommu_table *tbl, void *vaddr,
unsigned int npages, int direction)
{
- unsigned long entry, flags;
+ unsigned long entry;
dma_addr_t ret = bad_dma_address;
- spin_lock_irqsave(&tbl->it_lock, flags);
-
entry = iommu_range_alloc(tbl, npages);
if (unlikely(entry == bad_dma_address))
@@ -304,12 +313,9 @@ static dma_addr_t iommu_alloc(struct iommu_table *tbl, void *vaddr,
tce_build(tbl, entry, npages, (unsigned long)vaddr & PAGE_MASK,
direction);
- spin_unlock_irqrestore(&tbl->it_lock, flags);
-
return ret;
error:
- spin_unlock_irqrestore(&tbl->it_lock, flags);
printk(KERN_WARNING "Calgary: failed to allocate %u pages in "
"iommu %p\n", npages, tbl);
return bad_dma_address;
@@ -321,6 +327,7 @@ static void __iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
unsigned long entry;
unsigned long badbit;
unsigned long badend;
+ unsigned long flags;
/* were we called with bad_dma_address? */
badend = bad_dma_address + (EMERGENCY_PAGES * PAGE_SIZE);
@@ -337,6 +344,8 @@ static void __iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
tce_free(tbl, entry, npages);
+ spin_lock_irqsave(&tbl->it_lock, flags);
+
badbit = verify_bit_range(tbl->it_map, 1, entry, entry + npages);
if (badbit != ~0UL) {
if (printk_ratelimit())
@@ -346,18 +355,14 @@ static void __iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
}
__clear_bit_string(tbl->it_map, entry, npages);
+
+ spin_unlock_irqrestore(&tbl->it_lock, flags);
}
static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
unsigned int npages)
{
- unsigned long flags;
-
- spin_lock_irqsave(&tbl->it_lock, flags);
-
__iommu_free(tbl, dma_addr, npages);
-
- spin_unlock_irqrestore(&tbl->it_lock, flags);
}
static inline struct iommu_table *find_iommu_table(struct device *dev)
@@ -402,17 +407,12 @@ static void __calgary_unmap_sg(struct iommu_table *tbl,
void calgary_unmap_sg(struct device *dev, struct scatterlist *sglist,
int nelems, int direction)
{
- unsigned long flags;
struct iommu_table *tbl = find_iommu_table(dev);
if (!translate_phb(to_pci_dev(dev)))
return;
- spin_lock_irqsave(&tbl->it_lock, flags);
-
__calgary_unmap_sg(tbl, sglist, nelems, direction);
-
- spin_unlock_irqrestore(&tbl->it_lock, flags);
}
static int calgary_nontranslate_map_sg(struct device* dev,
@@ -433,7 +433,6 @@ int calgary_map_sg(struct device *dev, struct scatterlist *sg,
int nelems, int direction)
{
struct iommu_table *tbl = find_iommu_table(dev);
- unsigned long flags;
unsigned long vaddr;
unsigned int npages;
unsigned long entry;
@@ -442,8 +441,6 @@ int calgary_map_sg(struct device *dev, struct scatterlist *sg,
if (!translate_phb(to_pci_dev(dev)))
return calgary_nontranslate_map_sg(dev, sg, nelems, direction);
- spin_lock_irqsave(&tbl->it_lock, flags);
-
for (i = 0; i < nelems; i++ ) {
struct scatterlist *s = &sg[i];
BUG_ON(!s->page);
@@ -467,8 +464,6 @@ int calgary_map_sg(struct device *dev, struct scatterlist *sg,
s->dma_length = s->length;
}
- spin_unlock_irqrestore(&tbl->it_lock, flags);
-
return nelems;
error:
__calgary_unmap_sg(tbl, sg, nelems, direction);
@@ -476,7 +471,6 @@ error:
sg[i].dma_address = bad_dma_address;
sg[i].dma_length = 0;
}
- spin_unlock_irqrestore(&tbl->it_lock, flags);
return 0;
}